Comparison of Multiple Indirect Approaches to Estimate Streamflow in the Osage and Severn Rivers

Abstract

Accurate estimation of river flows is essential for effective water resource management. While direct flow measurements are ideal, they are often costly and time-consuming. Indirect techniques offer viable alternatives. This study compares three distinct indirect approaches—rating curves (RCs), hydraulic models, and data-driven models—to estimate streamflow in the Osage and Severn Rivers. The RC-based methods include the classic RC, flow area rating curve (FARC), and isovel contours–based RC. The isovel contours–based RC utilizes variables such as flow section area (A), water surface width (T), wetted perimeter (P), and velocity (U) derived from isovel contours. These variables also serve as inputs for data-driven models, including support vector regression (SVR), adaptive neural fuzzy inference system (ANFIS), and long short-term memory (LSTM) networks. The Conveyance Estimation System (CES), based on the Shiono and Knight Method (SKM), is used as a hydraulic model. A key innovation of this research is the integration of hydraulic variables into data-driven models and comparing them with other streamflow estimation approaches. Direct flow measurements, obtained using acoustic Doppler current profilers (ADCPs), were employed as a benchmark to evaluate the accuracy of the indirect methods. The results, assessed using metrics such as mean absolute error (MAE), root-mean-squared error (RMSE), Nash–Sutcliffe efficiency (NSE), Taylor diagrams, and box plots, reveal that the appropriate data-driven models (SVR and ANFIS) consistently outperform other approaches across both case studies. Integrating hydraulic variables into data-driven models, rather than using them with RCs, enhances flow estimation performance. Conversely, CES software performed poorly and showed significant deviations from other models in both rivers. However, no significant differences were found among the other studied models, suggesting that cost-effectiveness analysis should be applied to choose the best model. Accurately estimating river streamflow is essential for effective water resource management. Direct measurements can be challenging and time-consuming. Therefore, this study explores indirect methods that can offer reliable results. It compares three approaches: stage–discharge rating curves, hydraulic, and data-driven models. Based on historical measurements, stage–discharge rating curves are relationships between water depth and flow rate. Hydraulic models are computer simulations that consider river characteristics like channel shape and roughness to estimate flow. Using advanced statistical techniques, data-driven models use historically appropriate data to estimate future flows. By focusing on rivers like the Osage and Severn, the research highlights how data-driven models, such as machine learning models, outperform traditional methods in estimating streamflow. A key takeaway is that combining hydraulic parameters, like flow area, wetted perimeter, and velocity based on isovel contours, with machine learning models boosts accuracy. For engineers and managers, this means better estimation of river behavior, helping in efficient water resource planning.